The Potential Evaluation of Multisource Remote Sensing Data for Extracting Soil Moisture Based on the Method of BP Neural Network

The Soil Moisture Active Passive Validation Experiment 2012 (SMAPVEX12) conducted in Manitoba, Canada, over a 6-week period in 2012 provided high-quality data for soil moisture and vegetation, captured over a wide range of conditions coincident with airborne active and passive microwave acquisitions. In this study, the dataset measured in SMAPVEX12 was used as the input data of the backpropagation (BP) neural network to analyze the potential of different data sources or combination of data sources in the research of soil moisture inversion. By the analysis and comparison of the results, some valuable conclusions have been obtained: (1) passive microwave data has a stronger and more direct connection with the near-surface soil moisture (top 0 similar to 6 cm) than the active microwave sensing (radar backscatter coefficient) does, because the active microwave is more sensitive to the confounding effects of vegetation and surface roughness. (2) For the active microwave data, some combination form of the horizontal transmit and horizontal receive (HH) and vertical transmit and horizontal receive (VH) is a more rational selection than only HH to serve as one of the independent information sources for extracting the soil moisture. (3) The optical vegetation information provides the rich vegetation information that is indispensable for the microwave remote sensing data, especially for passive remote sensing, to extract the soil moisture with the average correlation coefficient increased by more than 0.05, at least, when adding the optical vegetation data. (4) Two kinds of forms of the optical vegetation information, either Vegetation Water Content (VWC) or Normalized Difference Water Index (NDWI), show no obvious differences to provide as auxiliary information for microwave remote sensing data in extracting soil moisture.